岩相及孔隙度预测在油气勘探中非常重要,为此,提出一种基于多阈值BIRCH(Balanced Iterative Reducing and Clustering Using Hierarchies)聚类的岩相预测方法,并以此为基础利用岭回归算法预测孔隙度。首先,根据地震波阻抗数据分布规律...岩相及孔隙度预测在油气勘探中非常重要,为此,提出一种基于多阈值BIRCH(Balanced Iterative Reducing and Clustering Using Hierarchies)聚类的岩相预测方法,并以此为基础利用岭回归算法预测孔隙度。首先,根据地震波阻抗数据分布规律启发式设定初始阈值,根据簇之间体积的不一致性,动态增加阈值,使用Agglomerative算法进行全局聚类以划分岩相;然后,以井点处孔隙度和地震波阻抗数据为输入,在同一岩相内采用改进的岭回归方法预测孔隙度。模型实验表明,多阈值BIRCH聚类方法具有良好的稳定性和较高的计算效率,岩相划分准确。实际数据结果表明,该方法能够准确预测孔隙度。展开更多
针对传统的特征选择算法只专注于特征间的相关性和冗余性而没有考虑特征之间交互作用的问题,提出一种基于交互信息的混合特征选择(hybrid feature selection based on mutual information,MIHFS)算法,该算法以K-最近邻算法的分类准确率...针对传统的特征选择算法只专注于特征间的相关性和冗余性而没有考虑特征之间交互作用的问题,提出一种基于交互信息的混合特征选择(hybrid feature selection based on mutual information,MIHFS)算法,该算法以K-最近邻算法的分类准确率作为衡量所选特征分类性能的评价指标,有效地去除了冗余和不相关的特征,保留了具有交互作用的特征。为了评估该算法的性能,从分类准确率、所选特征数量以及算法稳定性三方面,与最大相关最小冗余、联合互信息等7种特征选择算法在8个数据集上进行了实验比较和分析。实验结果表明:MIHFS算法具有较强的稳定性,不仅有效降低了特征空间的维数,而且在所选特征的分类性能方面明显优于其他特征选择算法。最后将MIHFS算法与灰色关联分析法-逼近理想解的排序技术法相结合并应用到高邮凹陷永安地区戴一段地质评价中,其评价结果准确率为80%,与实际钻探结果基本吻合,具有较高的可靠性,能够有效指导油气地质评价。展开更多
Azimuth gamma logging while drilling(LWD)is one of the important technologies of geosteering but the information of real-time data transmission is limited and the interpretation is difficult.This study proposes a meth...Azimuth gamma logging while drilling(LWD)is one of the important technologies of geosteering but the information of real-time data transmission is limited and the interpretation is difficult.This study proposes a method of applying artificial intelligence in the LWD data interpretation to enhance the accuracy and efficiency of real-time data processing.By examining formation response characteristics of azimuth gamma ray(GR)curve,the preliminary formation change position is detected based on wavelet transform modulus maxima(WTMM)method,then the dynamic threshold is determined,and a set of contour points describing the formation boundary is obtained.The classification recognition model based on the long short-term memory(LSTM)is designed to judge the true or false of stratum information described by the contour point set to enhance the accuracy of formation identification.Finally,relative dip angle is calculated by nonlinear least square method.Interpretation of azimuth gamma data and application of real-time data processing while drilling show that the method proposed can effectively and accurately determine the formation changes,improve the accuracy of formation dip interpretation,and meet the needs of real-time LWD geosteering.展开更多
文摘岩相及孔隙度预测在油气勘探中非常重要,为此,提出一种基于多阈值BIRCH(Balanced Iterative Reducing and Clustering Using Hierarchies)聚类的岩相预测方法,并以此为基础利用岭回归算法预测孔隙度。首先,根据地震波阻抗数据分布规律启发式设定初始阈值,根据簇之间体积的不一致性,动态增加阈值,使用Agglomerative算法进行全局聚类以划分岩相;然后,以井点处孔隙度和地震波阻抗数据为输入,在同一岩相内采用改进的岭回归方法预测孔隙度。模型实验表明,多阈值BIRCH聚类方法具有良好的稳定性和较高的计算效率,岩相划分准确。实际数据结果表明,该方法能够准确预测孔隙度。
文摘针对传统的特征选择算法只专注于特征间的相关性和冗余性而没有考虑特征之间交互作用的问题,提出一种基于交互信息的混合特征选择(hybrid feature selection based on mutual information,MIHFS)算法,该算法以K-最近邻算法的分类准确率作为衡量所选特征分类性能的评价指标,有效地去除了冗余和不相关的特征,保留了具有交互作用的特征。为了评估该算法的性能,从分类准确率、所选特征数量以及算法稳定性三方面,与最大相关最小冗余、联合互信息等7种特征选择算法在8个数据集上进行了实验比较和分析。实验结果表明:MIHFS算法具有较强的稳定性,不仅有效降低了特征空间的维数,而且在所选特征的分类性能方面明显优于其他特征选择算法。最后将MIHFS算法与灰色关联分析法-逼近理想解的排序技术法相结合并应用到高邮凹陷永安地区戴一段地质评价中,其评价结果准确率为80%,与实际钻探结果基本吻合,具有较高的可靠性,能够有效指导油气地质评价。
基金Supported by the PetroChina Major Scientific and Technological Project(ZD2019-183-006)Fundamental Scientific Research Fund of Central Universities(20CX05017A)China National Science and Technology Major Project(2016ZX05021-001)。
文摘Azimuth gamma logging while drilling(LWD)is one of the important technologies of geosteering but the information of real-time data transmission is limited and the interpretation is difficult.This study proposes a method of applying artificial intelligence in the LWD data interpretation to enhance the accuracy and efficiency of real-time data processing.By examining formation response characteristics of azimuth gamma ray(GR)curve,the preliminary formation change position is detected based on wavelet transform modulus maxima(WTMM)method,then the dynamic threshold is determined,and a set of contour points describing the formation boundary is obtained.The classification recognition model based on the long short-term memory(LSTM)is designed to judge the true or false of stratum information described by the contour point set to enhance the accuracy of formation identification.Finally,relative dip angle is calculated by nonlinear least square method.Interpretation of azimuth gamma data and application of real-time data processing while drilling show that the method proposed can effectively and accurately determine the formation changes,improve the accuracy of formation dip interpretation,and meet the needs of real-time LWD geosteering.
